Improving Machine Learning Performance

Improving machine learning model performance for predicting loan status.

Task : Classification

Requirements

• Python 3.7 or above
• Modules:
  • pandas
  • numpy
  • seaborn
  • matplotlib
  • scipy
  • sklearn
  • datetime
  • imblearn

Dataset

The dataset used in this experiment : (https://github.com/nzlul03/improving_ML_performance/blob/main/loan_approval.csv)

Steps to Improve Classification Model

• Feature Engineering
• Oversampling
• Hyper-parameter Tuning

Classification Model

In the main experiment (with the aim of improving model performance), we performed a comparison between the 2 models:

• Decision Tree
• Random Forest

We also tried to see the performance of other models (with default hyperparameter):

• Logistic Regression
• Support Vector Machine Linear (SVC)
• Gradient Boosting
• Ada Boost
• K-Nearest Neighbors
• Gaussian Naive Bayes
• Multinomial Naive Bayes
• LightGBM
• XGBoost
• Multi Layer Perceptron

Data Cleansing

• Handling duplicate values
• Handling missing values
• Handling outlier

Feature Encoding and Preprocessing

• Label encoder
• One hot encoder

The Result

• Correlation Matrix

• The Result of Experiment

• ROC Curve

• Feature Importance

The Other Experiment Results

ML Algo	AUC	Accuracy
Logistic Regression	64.01	59.32
SVC	62.08	43.22
Decision Tree	68.49	72.03
Random Forest	78.72	79.66
Gradient Boosting	79.39	78.81
Ada Boost	75.93	77.97
KNN	46.06	49.15
GaussianNB	64.9	64.41
MultinomialNB	42.31	41.53
LGBM	79.71	79.66
XGBoost	77.85	79.66
MLP	45.99	31.36